Interest point recommendation method and device based on association mode and heterogeneous bigraph

Abstract

The invention provides an interest point recommendation method and device based on an association mode and heterogeneous double graphs and a storage medium. The method comprises the steps that a user information set, an interest point information set, an interest point category information set and a user sign-in time information set are acquired; constructing an association mode according to the user information set, the interest point information set, the interest point category information set and the user sign-in time information set; respectively constructing a first heterogeneous graph and a second heterogeneous graph according to the association mode and the set; according to the association mode, based on vector embedding, respectively determining a first decomposition graph of the first heterogeneous graph and a second decomposition graph of the second heterogeneous graph; based on an information entropy weighting method, determining a first information entropy and a first weight corresponding to the first heterogeneous graph, and determining a second information entropy and a second weight corresponding to the second heterogeneous graph; and according to the first information entropy, the first weight, the second information entropy and the second weight, determining a recommendation score of each interest point. According to the method, the interest point recommendation accuracy can be improved.

Description

Technical Field

[0001] This document relates to the field of geographic information technology, and in particular to a method, device and storage medium for recommending points of interest based on association patterns and heterogeneous dual maps. Background Art

[0002] In recent years, the Point-of-Interest (POI) recommendation system based on location-based social networks (LBSN) has significantly improved the recommendation performance by integrating multi-dimensional contextual data such as user behavior, geographic information, time series and social relationships.

[0003] Compared to traditional product or movie recommendation systems, POI recommendations face more complex external dynamic factors and require analyzing the heterogeneous interactions between users and POIs. Existing research uses graph embedding techniques to model user-POI interaction networks and, to a certain extent, captures the relevance of multimodal data.

[0004] However, the existing POI recommendation system still has the following key problems, which lead to inaccurate recommendation results:

[0005] (1) Existing methods usually regard different contextual information as independent features of users and POIs, such as time, space, and semantics, and fail to fully explore the intrinsic coupling relationship between users and POIs based on features.

[0006] (2) The current node information aggregation method based on path instances directly uses the attention mechanism to assign weights, which fails to quantify the differences in information contributions of different association patterns. Summary of the Invention

[0007] In view of the above solutions, the present application aims to propose a method, device and storage medium for recommending points of interest based on association patterns and heterogeneous dual graphs to solve at least one of the above problems.

[0008] In a first aspect, one or more embodiments of the present application provide a method for recommending points of interest based on association patterns and heterogeneous dual graphs, including:

[0009] Obtain user information set, point of interest information set, point of interest category information set and user check-in time information set;

[0010] constructing at least one association pattern based on the user information set, the POI information set, the POI category information set, and the user check-in time information set, wherein the association pattern is used to represent an association relationship between the user information, the POI information, the POI category information, and the user check-in time information;

[0011] Constructing a first heterogeneous graph of interest information based on semantic relationships and a second heterogeneous graph of user information based on time association, respectively, according to the association pattern, the user information set, the interest point information set, the interest point category information set, and the user check-in time information set;

[0012] Determining, according to the association pattern and based on vector embedding, a first decomposition graph of the first heterogeneous graph and a second decomposition graph of the second heterogeneous graph;

[0013] Determining a first information entropy and a first weight corresponding to the first heterogeneous graph based on an information entropy weighting method;

[0014] Determining a second information entropy and a second weight corresponding to the second heterogeneous graph based on an information entropy weighting method;

[0015] A recommendation score for each point of interest is determined based on the first information entropy, the first weight, the second information entropy, and the second weight.

[0016] Furthermore, after obtaining the decomposition graph and before embedding the vector, the method further includes:

[0017] Negative samples are added to the first decomposition graph or the second decomposition graph.

[0018] Furthermore, adding negative samples includes:

[0019] Set the total number of training rounds, the initial ratio of negative samples, and the maximum ratio of negative samples;

[0020] For each training round, the current negative sample ratio is adjusted based on the linear rule and the initial negative sample ratio until the current negative sample ratio is equal to the maximum negative sample ratio.

[0021] Furthermore, adjusting the current negative sample ratio based on the linear rule and the initial negative sample ratio includes:

[0022] The current negative sample ratio is calculated using the following formula:

[0023]

[0024] Among them, cr is the current negative sample ratio, n is the total number of training rounds, c is the current training round, ω1 is the initial ratio and ω2 is the maximum ratio.

[0025] Furthermore, based on an information entropy weighting method, determining a first information entropy and a first weight corresponding to the first heterogeneous graph includes:

[0026] Determine the information entropy of the interaction between the user and the point of interest in the first decomposition graph based on the information entropy weighting method;

[0027] Determining the connection information entropy corresponding to the first decomposition graph based on an information entropy weighting method;

[0028] Constructing an information entropy matrix corresponding to the first decomposition graph according to the interaction information entropy and the connection information entropy;

[0029] According to the information entropy matrix corresponding to the first decomposition graph, a first information entropy and a first weight corresponding to the first heterogeneous graph are determined.

[0030] Further, determining a first weight corresponding to the first heterogeneous graph according to the information entropy matrix corresponding to the first decomposition graph includes:

[0031] Based on the attention mechanism, the importance coefficient of the information entropy matrix is ​​calculated;

[0032] According to the importance coefficient of the information entropy matrix, the importance coefficient corresponding to the first heterogeneous graph is calculated.

[0033] Furthermore, based on the information entropy weighting method, determining the second information entropy and the second weight corresponding to the second heterogeneous graph includes:

[0034] Determine the information entropy of the interaction between the user and the point of interest in the second decomposition graph based on the information entropy weighting method;

[0035] Determining the connection information entropy corresponding to the second decomposition graph based on an information entropy weighting method;

[0036] Constructing an information entropy matrix corresponding to the second decomposition graph according to the interaction information entropy and the connection information entropy;

[0037] According to the information entropy matrix corresponding to the second decomposition graph, a second information entropy and a second weight corresponding to the second heterogeneous graph are determined.

[0038] Furthermore, it is characterized in that

[0039] Determining a second weight corresponding to the second heterogeneous graph according to the information entropy matrix corresponding to the second decomposition graph includes:

[0040] Based on the attention mechanism, the importance coefficient of the information entropy matrix is ​​calculated;

[0041] According to the importance coefficient of the information entropy matrix, the importance coefficient corresponding to the second heterogeneous graph is calculated.

[0042] In a second aspect, one or more embodiments of the present application provide a device for recommending points of interest based on association patterns and heterogeneous dual graphs, including:

[0043] An acquisition module, used to acquire a user information set, a point of interest information set, a point of interest category information set, and a user check-in time information set;

[0044] a construction module, configured to construct at least one association pattern based on the user information set, the POI information set, the POI category information set, and the user check-in time information set, wherein the association pattern is used to characterize the association relationship between the user information, the POI information, the POI category information, and the user check-in time information; and to construct a first heterogeneous graph of POI information based on semantic relationships and a second heterogeneous graph of user information based on time associations based on the user information set, the POI information set, the POI category information set, and the user check-in time information set, respectively;

[0045] A data processing module is used to determine, according to the association pattern and based on vector embedding, a first decomposition graph of the first heterogeneous graph and a second decomposition graph of the second heterogeneous graph; based on an information entropy weighted method, determine a first information entropy and a first weight corresponding to the first heterogeneous graph; based on an information entropy weighted method, determine a second information entropy and a second weight corresponding to the second heterogeneous graph; and determine a recommendation score for each point of interest based on the first information entropy, the first weight, the second information entropy, and the second weight.

[0046] In a third aspect, an embodiment of the present application provides a storage medium for storing computer-executable instructions, characterized in that when the computer-executable instructions are executed, the steps of the method for recommending points of interest based on association patterns and heterogeneous dual graphs described in any one of the first aspects are implemented.

[0047] Compared with the existing technology, this application can at least achieve the following technical effects:

[0048] This application first constructs a heterogeneous dual graph of points of interest and users, laying the foundation for the subsequent fusion of multiple factors. Secondly, it constructs an association pattern, and transforms the heterogeneous graph into a decomposition graph based on the association pattern to achieve parameter fusion from multiple dimensions. Thirdly, based on the decomposition graph, vector embedding and information entropy weighting method, the weight of the heterogeneous graph is determined. Finally, based on the weight of the heterogeneous graph, the recommendation score of each point of interest is determined. It can be seen that this application can fully explore the intrinsic coupling relationship between users and points of interest based on features to improve the accuracy of the recommendation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate one or more embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0050] Figure 1 A flowchart of a method for recommending points of interest based on association patterns and heterogeneous dual graphs provided in one or more embodiments of the present application;

[0051] Figure 2 A schematic structural diagram of a device for recommending points of interest based on association patterns and heterogeneous dual graphs according to one or more embodiments of the present application. DETAILED DESCRIPTION

[0052] In order to enable those skilled in the art to better understand the technical solutions in one or more embodiments of the present application, the technical solutions in one or more embodiments of the present application will be clearly and completely described below in conjunction with the drawings in one or more embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on one or more embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without making creative work should fall within the scope of protection of this document.

[0053] The embodiment of the present application provides a method for recommending points of interest based on association patterns and heterogeneous dual graphs, such as Figure 1 As shown, the following steps are included:

[0054] Step 1: Obtain user information set, point of interest information set, point of interest category information set, and user check-in time information set.

[0055] In the embodiment of the present application, the user set is defined as U = {u1, u2, ..., u M}, the set of points of interest is P = {p1,p2,……,p N}, interest point category set C = {c1, c2, ..., c O}, the user check-in time set is T = {t1, t2, ..., t I}, where M, N, and O represent the number of users, the number of points of interest, and the number of categories, respectively. I represents the number of time periods, and a day is divided into 24 time periods.

[0056] Step 2: Construct at least one association pattern based on the user information set, the point of interest information set, the point of interest category information set, and the user check-in time information set.

[0057] In the embodiment of the present application, the association pattern is used to characterize the association relationship between user information, point of interest information, point of interest category information and user check-in time information. Specifically, the association pattern is defined as:

[0058] If nodes v1 and v2 are connected through another node of type R, then there is an indirect association pattern (v1, v2, R) between v1 and v2. If v1 and v2 are connected through an edge of type E, then there is a direct association pattern (v1, v2, E) between v1 and v2. For example, in the relationship u1-t1-u2, nodes u1 and u2 have a time association pattern; in the relationship u3-p1, nodes u3 and p1 have a direct check-in association pattern.

[0059] The specific association mode is:

[0060] The direct check-in association patterns (U, P, Check-in) and (P, U, Check-in) represent the interaction information between users and POIs, where Check-in indicates that users check in at a POI. The indirect association pattern (U, U, P) indicates that users establish connections through commonly visited POIs, capturing the similarity or social preferences between users. The indirect association pattern (P, P, U) indicates that POIs are connected through commonly visited users, reflecting to a certain extent the possible geographical proximity between POIs. The indirect association pattern (P, P, C) indicates that connections are established through the category information of POIs, reflecting the semantic and functional similarities of POIs. The indirect association pattern (U, U, T) indicates that users establish connections through common visit times, capturing the similarity of behavioral patterns between users.

[0061] Step 3: Based on the association pattern, user information set, point of interest information set, point of interest category information set and user check-in time information set, a first heterogeneous graph of point of interest information based on semantic relationships and a second heterogeneous graph of user information based on time association are constructed respectively.

[0062] In the embodiment of the present application, the first heterogeneous graph construction process is:

[0063] Construct a heterogeneous graph G based on semantic relations of interest points C ={U,P,C,E UP ,E PC}, where the edge connecting the user and the point of interest is represented by E UP ={(u m ,p n )|u m ∈U,p n ∈P}, the edge connecting the interest point and its category is represented by E PC ={(p n ,c o )|p n ∈P,c o∈C}. In the original dataset, the points of interest only have name attributes. In order to obtain the nodes of interest band categories in the heterogeneous graph, this application uses a pre-trained word embedding model to calculate the word vectors of all the names of all points of interest, and then uses the K-means algorithm to cluster the word vectors. According to the classification and grading standards of urban infrastructure construction, all the points of interest categories are unsupervisedly classified into 14 clusters to obtain the high-order prototype vector E of each cluster. c , the calculation formula is:

[0064]

[0065] where x i ,y i are sample word vectors and their category labels, respectively. is the prototype vector of the kth category of the interest point, S k is the support sample set of category k, f φ (·) is the embedded neural network. First, the direct check-in association pattern (P, U, Check-in) is used to extract nodes from the semantic relationship graph to obtain the interest point-user interaction decomposition graph G centered on the interest point. inter_P , is used to mine the explicit interaction information between users and points of interest, and the edge weight is set to the average normalized value of the number of user check-ins; then the same method is used to use the indirect association patterns (P, P, U) and (P, P, C) to obtain the point of interest isomorphism decomposition graph G homo_PU With G homo_PC , is used to mine implicit matching relationships between interest points, and the edge weight is set to the average normalized value of the local distance between interest points. Other association patterns can also refer to the above method to obtain heterogeneous graphs.

[0066] The second heterogeneous graph construction process is:

[0067] Construct a heterogeneous graph G of users based on time association T ={P,U,T,E PU ,E UT}, where the edge connecting the point of interest and the user is represented by E PU ={(p m ,u n )|p m ∈P,u n ∈U}, the edge connecting the user and the check-in time is represented by E UT ={(u n ,t i )|u n ∈U,t i∈T}. In this embodiment of the present invention, a day is divided into 24-hour time periods, and a unique ID value is assigned to each time period for one-hot encoding as the feature of the time node. Then, the ID value of the time node is mapped to the implicit vector space by embedding to obtain the initial vector representation. The calculation formula is: E t =M T e t , where e t is the time node one-hot vector, M T is the weight conversion matrix, E t is the time feature vector. First, the direct check-in association pattern (P, U, Check-in) is used to extract nodes from the time association graph to obtain the user-interest point interaction decomposition graph G centered on the user inter_U , is used to mine the explicit interaction information between users and points of interest, and the edge weight is set to the average normalized value of the number of check-ins; then the user isomorphic decomposition graph G is obtained by using the indirect association patterns (U, U, P) and (U, U, T) in the same way homo_UP With G homo_UT , used to mine implicit user associations. Other association patterns can also refer to the above method to obtain heterogeneous graphs. The edge weight calculation formula is:

[0068]

[0069] Among them, N(u i ) and N(u j ) are u i and u j Neighbor set, lg(·) is the logarithmic operation, |N(u k )| is u k If u i and u j If they visit a common point of interest, they are considered neighbors.

[0070] Step 4: According to the association pattern and based on vector embedding, a first decomposition graph of the first isomeric graph and a second decomposition graph of the second isomeric graph are determined respectively.

[0071] In the embodiment of the present application, the method for determining the first exploded view and the second exploded view is specifically as follows:

[0072] Since users check in at multiple time periods, simple vector addition may not be able to learn the nonlinear characteristics of users’ time preferences. i or point of interest p j Node, in order to capture the pattern of user behavior changing over time, for user u i The number of check-ins in each time period is normalized, and the normalized value is used as the weight to weight each time feature vector.

[0073] For example, if a user checks in 10 times in the entire dataset—2 in the morning, 6 at noon, and 2 in the evening—the user likely prefers midday activity. The calculated weights are 0.2 for morning, 0.6 for noon, and 0.2 for evening. Morning, noon, and evening each have a time feature vector. These are multiplied by the weights and then added together to create a fused time feature vector. This approach reduces the number of input vectors required in subsequent calculations, thereby reducing computational overhead.

[0074] The weighted time feature vector and the user random embedding vector are then concatenated and input into a two-layer MLP to obtain the time-aware user feature vector.

[0075] The category attribute of the interest point is one of the bases for model decision making. In order to increase the interpretability of the model, the interest point p j The random embedding vector and category embedding vector of are input into a two-layer MLP to obtain category-aware POI embedding, so as to recommend certain POIs according to the user's category preferences.

[0076] Correlation pattern ξ m Node v below i ={u i ∨p i |u i ,p i ∈U∪P} and its neighboring nodes form a decomposition graph based on a specific pattern, and then add nodes from v to the decomposition graph at a certain ratio ω i The difficult negative samples selected from the unchecked nodes are used to obtain a decomposition graph with new sample composition. All node features are represented as The edge weights of negative samples are all set to -1, and then all edge weights are re-normalized by the mean. The purpose of using mean normalization is to retain the negative weight information of negative samples and improve the sample discrimination ability of the model.

[0077] In order to obtain the feature embedding vectors of users and interest points in the current association mode, the graph attention network is used to aggregate node neighborhood information and add its own features in the calculation. i ,v j >, the attention coefficient calculation formula is:

[0078]

[0079] Among them, e ij Represents node v j Relative to v i The attention coefficient, and W are the parameters of shared learning, σ is the activation function, and || is the vector concatenation operation.

[0080] Then use the softmax function to normalize the attention coefficient, and the calculation formula is:

[0081]

[0082] Finally update the node characteristics:

[0083]

[0084] in, Indicates v i The features after integrating neighborhood information, σ is the activation function.

[0085] Step 5: According to the first decomposition graph, based on the information entropy weighting method, determine the first information entropy and the first weight corresponding to the first heterogeneous graph.

[0086] In the embodiment of the present application, the embedding representations of multiple users and points of interest can be obtained according to different association mode types, so the final user and point of interest node embedding needs to be further fused. Under all association modes, L embedding vectors of users and points of interest can be obtained, which are expressed as Use these vectors to construct the user embedding matrix: M user ∈R M×d ,

[0087] Use these vectors to construct the interest point embedding matrix: M poi ∈R N×d , d is the embedding dimension.

[0088] Then, based on these user embedding matrices and point of interest embedding matrices, we use the information entropy weighting method to evaluate the importance of the preferred feature embeddings under different association patterns. Afterwards, we determine the weights of the heterogeneous graph based on these importances.

[0089] Step 6: According to the second decomposition graph, based on the information entropy weighting method, determine the second information entropy and the second weight corresponding to the second heterogeneous graph.

[0090] In the embodiment of the present application, the method for determining the second information entropy and the second weight is the same as the method for determining the first information entropy and the first weight.

[0091] Step 7: Determine the recommendation score of each point of interest based on the first information entropy, the first weight, the second information entropy, and the second weight.

[0092] In the embodiment of the present application, the specific process of determining the recommendation score is as follows:

[0093] The inner product operation of the final user embedding vector and the interest point embedding vector is used to obtain the score, which is the user's recommendation score for each interest point. The calculation formula is:

[0094] Score = z i user ·z j poi ;

[0095] Then, the probability distribution of users visiting points of interest is obtained through MLP. The optimization function selects the root mean square error loss L, and the calculation formula is:

[0096]

[0097] where y n is the true value, y n =1 means the user has visited the point of interest, y n =0 means not visited, represents the predicted probability of a user visiting the nth point of interest.

[0098] In the embodiment of the present application, the model only uses positive samples from the check-in data during training, which will ignore the implicit negative feedback information of users who have not checked in to the points of interest. Adding negative samples can better learn the characteristics of user interest preferences. However, the ratio of the number of positive samples to negative samples seriously affects the model performance and recommendation effect. In some methods, the proportion of positive samples in the total samples is fixed and negative samples are randomly selected. This may lead to problems such as model overfitting, insufficient generalization ability, and sample imbalance. Therefore, the present application designs a method for linearly adaptively adjusting the composition of training samples based on evaluation indicators. Specifically, the total number of training rounds, the initial proportion of negative samples, and the maximum proportion of negative samples are set;

[0099] For each training round, the current negative sample ratio is adjusted based on the linear rule and the initial negative sample ratio until the current negative sample ratio is equal to the maximum negative sample ratio.

[0100] The current negative sample ratio is calculated using the following formula:

[0101]

[0102] Among them, cr is the current negative sample ratio, n is the total number of training rounds, c is the current training round, ω1 is the initial ratio and ω2 is the maximum ratio.

[0103] Since there are other nodes in the negative sample that have different relevance to the current node, in order to enhance the model's ability to distinguish samples, the cosine similarity is used to calculate the difficult negative sample with the closest feature to the current node and add it to the decomposition graph. The cosine similarity calculation formula is:

[0104]

[0105] Here, ||·|| represents the modulo operation of the node feature vector. The rate of change of the training loss can significantly reflect the model's convergence performance. Therefore, the loss value is used as the evaluation criterion for adjusting the sample composition. If the average loss calculated from the x-round loss values ​​is not greater than the threshold ε, the loss value is considered to be declining slowly, and the proportion of difficult negative samples is adjusted accordingly.

[0106] In the embodiment of the present application, based on the information entropy weighting method, the first information entropy and the first weight corresponding to the first heterogeneous graph are determined. The specific process is:

[0107] Based on the information entropy weighting method, the interaction information entropy between users and points of interest in the first decomposition graph is determined; based on the information entropy weighting method, the connection information entropy corresponding to the first decomposition graph is determined; based on the interaction information entropy and the connection information entropy, the information entropy matrix corresponding to the first decomposition graph is constructed; based on the information entropy matrix corresponding to the first decomposition graph, the first information entropy and the first weight corresponding to the first heterogeneous graph are determined.

[0108] In an embodiment of the present application, the first weight corresponding to the first heterogeneous graph is determined based on the information entropy matrix corresponding to the first decomposition graph. The specific process is: based on the attention mechanism, the importance coefficient of the information entropy matrix is ​​calculated; based on the importance coefficient of the information entropy matrix, the importance coefficient corresponding to the first heterogeneous graph is calculated.

[0109] In the embodiment of the present application, based on the information entropy weighting method, the second information entropy and the second weight corresponding to the second heterogeneous graph are determined. The specific process is:

[0110] Based on the information entropy weighting method, the interaction information entropy between users and points of interest in the second decomposition graph is determined; based on the information entropy weighting method, the connection information entropy corresponding to the second decomposition graph is determined; according to the interaction information entropy and the connection information entropy, the information entropy matrix corresponding to the second decomposition graph is constructed; according to the information entropy matrix corresponding to the second decomposition graph, the second information entropy and the second weight corresponding to the second heterogeneous graph are determined.

[0111] In the embodiment of the present application, the second weight corresponding to the second heterogeneous graph is determined according to the information entropy matrix corresponding to the second decomposition graph. The specific process is:

[0112] Based on the attention mechanism, the importance coefficient of the information entropy matrix is ​​calculated; according to the importance coefficient of the information entropy matrix, the importance coefficient corresponding to the second heterogeneous graph is calculated.

[0113] It should be noted that the method for determining the first information entropy and the first weight is the same as the method for determining the second information entropy and the second weight. For ease of explanation, only one example is used here for illustration. Specifically,

[0114] First, we calculate the interaction information entropy between users and points of interest in the decomposition graph under different association modes, and the connection information entropy between users and users, and between points of interest. The calculation formulas are:

[0115]

[0116]

[0117] in, For node v i The number of neighbor nodes, v i degree, lg(·) is the logarithmic operation, is the association mode ξ m The node interaction matrix under , the element value is the number of interactions, if the number of interactions is 0, the entropy value is 0.

[0118] Under m association modes, all node entropy values ​​are obtained to form the information entropy matrix M etp ∈R m×d , d∈{M,N} is the embedding dimension. Then a layer of MLP is used to perform a nonlinear transformation on the information entropy matrix, the formula is:

[0119] M 'etp =tanh(W etp ·M etp +b etp );

[0120] W etp ∈R m×d is the learnable weight matrix, b etp ∈R d is the bias term. Finally, the attention mechanism is introduced to adjust the importance of nodes for different association patterns. The calculation formula is:

[0121]

[0122] Where d' is the hidden layer dimension, W att ∈R d′×d is the learnable weight matrix, b att ∈R d′ is the bias vector, a∈R d′ is a parameter vector, and the weights are normalized by the softmax function to obtain the importance coefficient of each association mode, which is calculated as follows:

[0123]

[0124] Where m is the number of association patterns.

[0125] By weighting the embedding representations of users and points of interest using the importance coefficients, we can obtain the user embedding vector and the point of interest embedding vector. The calculation method is:

[0126]

[0127] Among them, M i user is the embedding matrix for the i-th user, M j poi Embedding matrix for the jth interest point.

[0128] The embodiment of the present application provides a device for recommending points of interest based on association patterns and heterogeneous dual graphs, such as Figure 2 As shown, including:

[0129] Acquisition module 201, for acquiring user information set, point of interest information set, point of interest category information set and user check-in time information set;

[0130] A construction module 202 is configured to construct at least one association pattern based on the user information set, the POI information set, the POI category information set, and the user check-in time information set, wherein the association pattern is used to represent the association relationship between the user information, the POI information, the POI category information, and the user check-in time information; and to construct a first heterogeneous graph of POI information based on semantic relationships and a second heterogeneous graph of user information based on time association based on the user information set, the POI information set, the POI category information set, and the user check-in time information set, respectively;

[0131] The data processing module 203 is used to determine the first decomposition graph of the first heterogeneous graph and the second decomposition graph of the second heterogeneous graph respectively according to the association pattern and based on vector embedding; determine the first information entropy and first weight corresponding to the first heterogeneous graph based on the information entropy weighting method; determine the second information entropy and second weight corresponding to the second heterogeneous graph based on the information entropy weighting method; and determine the recommendation score of each point of interest based on the first information entropy, the first weight, the second information entropy and the second weight.

[0132] An embodiment of the present application provides a storage medium for storing computer-executable instructions, characterized in that when the computer-executable instructions are executed, the steps of the method for recommending points of interest based on association patterns and heterogeneous dual graphs described in any one of the embodiments are implemented.

[0133] It should be noted that the embodiment of the storage medium in this application and the embodiment of the method for recommending points of interest based on association patterns and heterogeneous dual graphs in this application are based on the same inventive concept. Therefore, the specific implementation of this embodiment can refer to the aforementioned corresponding implementation of the method for recommending points of interest based on association patterns and heterogeneous dual graphs, and the repeated parts will not be repeated.

[0134] The foregoing description describes specific embodiments of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0135] In the 1930s, technological improvements could be clearly distinguished as either hardware improvements (for example, improvements to circuit structures like diodes, transistors, and switches) or software improvements (improvements to process flows). However, with the advancement of technology, many process flow improvements today can now be considered direct improvements to hardware circuit structures. Designers almost always create the corresponding hardware circuit structure by programming the improved process flow into the hardware circuit. Therefore, it cannot be said that a process flow improvement cannot be implemented using hardware modules. For example, a programmable logic device (PLD), such as a field programmable gate array (FPGA), is an integrated circuit whose logical function is determined by user programming. Designers can "integrate" a digital system on a PLD by programming it themselves, without having to hire a chip manufacturer to design and manufacture a dedicated integrated circuit chip. Moreover, nowadays, instead of manually fabricating integrated circuit chips, this programming is mostly done using "logic compiler" software. This is similar to the software compiler used when developing programs. Before compilation, the original code must also be written in a specific programming language, called a hardware description language (HDL). There is not just one HDL, but many, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art will also understand that by simply programming the method flow in one of these hardware description languages ​​and then programming it into an integrated circuit, a hardware circuit that implements the logic method flow can be easily obtained.

[0136] The controller can be implemented in any suitable manner. For example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also know that in addition to implementing the controller in a purely computer-readable program code format, the controller can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, such a controller can be considered a hardware component, and the devices included therein for implementing various functions can also be considered as structures within the hardware component. Or even, the devices for implementing various functions can be considered as both software modules that implement the method and structures within the hardware component.

[0137] The systems, devices, modules, or units described in the above embodiments may be implemented by computer chips or entities, or by products having certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0138] For the convenience of description, the above devices are described as being divided into various units according to their functions. Of course, when implementing the embodiments of the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0139] Those skilled in the art will appreciate that one or more embodiments of the present application may be provided as a method, system, or computer program product. Therefore, one or more embodiments of the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0140] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0141] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0142] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0143] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0144] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0145] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0146] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0147] One or more embodiments of the present application may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. One or more embodiments of the present application may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communications network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0148] The various embodiments in this application are described in a progressive manner. Similar parts between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the system embodiment is generally similar to the method embodiment, so the description is relatively simple. For relevant parts, refer to the partial description of the method embodiment.

[0149] The foregoing description is merely an example of the present invention and is not intended to limit the present invention. Persons skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be included within the scope of the claims herein.

Claims

1. A method for recommending points of interest based on association patterns and heterogeneous dual graphs, characterized in that: include: Obtain user information set, point of interest information set, point of interest category information set and user check-in time information set; constructing at least one association pattern based on the user information set, the POI information set, the POI category information set, and the user check-in time information set, wherein the association pattern is used to represent an association relationship between the user information, the POI information, the POI category information, and the user check-in time information; Constructing a first heterogeneous graph of interest information based on semantic relationships and a second heterogeneous graph of user information based on time association, respectively, according to the association pattern, the user information set, the interest point information set, the interest point category information set, and the user check-in time information set; Determining, according to the association pattern and based on vector embedding, a first decomposition graph of the first heterogeneous graph and a second decomposition graph of the second heterogeneous graph; Determining, based on the information entropy weighting method, a first information entropy and a first weight corresponding to the first heterogeneous graph according to the first decomposition graph; Determine, based on the information entropy weighting method, a second information entropy and a second weight corresponding to the second heterogeneous graph according to the second decomposition graph; A recommendation score for each point of interest is determined based on the first information entropy, the first weight, the second information entropy, and the second weight.

2. The method according to claim 1, characterized in that After obtaining the decomposition graph and before embedding the vector, the method further includes: Negative samples are added to the first decomposition graph or the second decomposition graph.

3. The method according to claim 2, characterized in that Adding negative samples includes: Set the total number of training rounds, the initial ratio of negative samples, and the maximum ratio of negative samples; For each training round, the current negative sample ratio is adjusted based on the linear rule and the initial negative sample ratio until the current negative sample ratio is equal to the maximum negative sample ratio.

4. The method according to claim 3, characterized in that Adjusting the current negative sample ratio based on the linear rule and the initial negative sample ratio includes: The current negative sample ratio is calculated using the following formula: Among them, cr is the current negative sample ratio, n is the total number of training rounds, c is the current training round, ω1 is the initial ratio and ω2 is the maximum ratio.

5. The method according to claim 1, wherein Determining a first information entropy and a first weight corresponding to the first heterogeneous graph based on the first decomposition graph based on an information entropy weighting method includes: Determine the information entropy of the interaction between the user and the point of interest in the first decomposition graph based on the information entropy weighting method; Determining the connection information entropy corresponding to the first decomposition graph based on an information entropy weighting method; Constructing an information entropy matrix corresponding to the first decomposition graph according to the interaction information entropy and the connection information entropy; According to the information entropy matrix corresponding to the first decomposition graph, a first information entropy and a first weight corresponding to the first heterogeneous graph are determined.

6. The method according to claim 5, characterized in that Determining a first weight corresponding to the first heterogeneous graph according to an information entropy matrix corresponding to the first decomposition graph includes: Based on the attention mechanism, the importance coefficient of the information entropy matrix is ​​calculated; According to the importance coefficient of the information entropy matrix, the importance coefficient corresponding to the first heterogeneous graph is calculated.

7. The method according to claim 1, characterized in that Determining, based on an information entropy weighting method and according to the second decomposition graph, a second information entropy and a second weight corresponding to the second heterogeneous graph, including: Determine the information entropy of the interaction between the user and the point of interest in the second decomposition graph based on the information entropy weighting method; Determining the connection information entropy corresponding to the second decomposition graph based on an information entropy weighting method; Constructing an information entropy matrix corresponding to the second decomposition graph according to the interaction information entropy and the connection information entropy; According to the information entropy matrix corresponding to the second decomposition graph, a second information entropy and a second weight corresponding to the second heterogeneous graph are determined.

8. The method for recommending points of interest based on association patterns and heterogeneous dual graphs according to claim 7, characterized in that: Determining a second weight corresponding to the second heterogeneous graph according to the information entropy matrix corresponding to the second decomposition graph includes: Based on the attention mechanism, the importance coefficient of the information entropy matrix is ​​calculated; According to the importance coefficient of the information entropy matrix, the importance coefficient corresponding to the second heterogeneous graph is calculated.

9. A point of interest recommendation device based on association patterns and heterogeneous dual graphs, characterized in that: include: An acquisition module, used to acquire a user information set, a point of interest information set, a point of interest category information set, and a user check-in time information set; a construction module, configured to construct at least one association pattern based on the user information set, the POI information set, the POI category information set, and the user check-in time information set, wherein the association pattern is used to characterize the association relationship between the user information, the POI information, the POI category information, and the user check-in time information; and to construct a first heterogeneous graph of POI information based on semantic relationships and a second heterogeneous graph of user information based on time associations based on the user information set, the POI information set, the POI category information set, and the user check-in time information set, respectively; a data processing module, configured to determine, according to the association pattern and based on vector embedding, a first decomposition graph of the first heterogeneous graph and a second decomposition graph of the second heterogeneous graph; Determining a first information entropy and a first weight corresponding to the first heterogeneous graph based on an information entropy weighting method; Determining a second information entropy and a second weight corresponding to the second heterogeneous graph based on an information entropy weighting method; A recommendation score for each point of interest is determined based on the first information entropy, the first weight, the second information entropy, and the second weight.

10. A storage medium for storing computer-executable instructions, characterized in that: When the computer executable instructions are executed, the steps of the method for recommending points of interest based on association patterns and heterogeneous dual graphs according to any one of claims 1 to 8 are implemented.

Images